pixart-900m-1024-ft

This is a full rank finetune derived from ptx0/pixart-900m-1024-ft-large.

The main validation prompt used during training was:

ethnographic photography of teddy bear at a picnic, ears tucked behind a cozy hoodie looking darkly off to the stormy picnic skies

Validation settings

• CFG: 4.5
• CFG Rescale: 0.0
• Steps: 25
• Sampler: None
• Seed: 42
• Resolutions: 1024x1024,1344x768,916x1152

Note: The validation settings are not necessarily the same as the training settings.

You can find some example images in the following gallery:

The text encoder was not trained. You may reuse the base model text encoder for inference.

Training settings

• Training epochs: 7
• Training steps: 100000
• Learning rate: 1e-06
• Effective batch size: 192
  • Micro-batch size: 24
  • Gradient accumulation steps: 1
  • Number of GPUs: 8
• Prediction type: epsilon
• Rescaled betas zero SNR: False
• Optimizer: AdamW, stochastic bf16
• Precision: Pure BF16
• Xformers: Not used

Datasets

photo-concept-bucket

• Repeats: 0
• Total number of images: ~567552
• Total number of aspect buckets: 1
• Resolution: 1.0 megapixels
• Cropped: True
• Crop style: random
• Crop aspect: square

Inference

import torch
from diffusers import DiffusionPipeline




model_id = 'pixart-900m-1024-ft'
prompt = 'ethnographic photography of teddy bear at a picnic, ears tucked behind a cozy hoodie looking darkly off to the stormy picnic skies'
negative_prompt = 'blurry, cropped, ugly'
pipeline = DiffusionPipeline.from_pretrained(model_id)
pipeline.to('cuda' if torch.cuda.is_available() else 'mps' if torch.backends.mps.is_available() else 'cpu')

prompt = "ethnographic photography of teddy bear at a picnic, ears tucked behind a cozy hoodie looking darkly off to the stormy picnic skies"
negative_prompt = "blurry, cropped, ugly"

pipeline = DiffusionPipeline.from_pretrained(model_id)
pipeline.to('cuda' if torch.cuda.is_available() else 'mps' if torch.backends.mps.is_available() else 'cpu')
image = pipeline(
    prompt=prompt,
    negative_prompt='blurry, cropped, ugly',
    num_inference_steps=25,
    generator=torch.Generator(device='cuda' if torch.cuda.is_available() else 'mps' if torch.backends.mps.is_available() else 'cpu').manual_seed(1641421826),
    width=1152,
    height=768,
    guidance_scale=4.5,
    guidance_rescale=0.0,
).images[0]
image.save("output.png", format="PNG")